An external hard drive is very helpful so that you can do sequential reads and writes without your heads skipping around too much. If you can isolate your audio to a separate drive, you can do the math.
A "typical" 7200 RPM drive can read about 30-40 megabytes per second (depends on platter size, block size, cache size, length of test, etc.). We'll go with the lower of these two numbers for simplicity. It can write about 10-15 megabytes per second (We'll go with the higher of these two numbers for simplicity; you'll see why later). It cannot, however, maintain the read transfer rate while writing at the same time. This dramatically lowers your throughput when you're recording while playing back. To simplify, since we're already rounding down to rather low numbers vs. peak sustained throughput, let's take seek time out of the equation entirely, and assume that time spent writing is not time spent reading, and that the fractions work out evenly.
24 bits per sample * 96,000 samples per second = 2,304,000 bits per second. 2,250 kilobits per second. 281.25 kilobytes per second. Or roughly 0.275 megabytes per second.
Frequently, you're recording stereo. I'll assume you're the "typical" home-studio recording artist, and have a maximum of two incoming channels at a time. That's 562.5 kilobytes per second, or roughtly half a megabyte per second.
The hard drive can do sustained sequential writes at roughly 15 megabytes per second. No problem so far recording a single track.
Now assume we've recorded two stereo tracks. That's roughly 1 megabyte per second. Once again, in sequential reads, the hard disk does just fine on playback. That's only 1/30th of its max sequential read speed.
Unfortunately, though, this *isn't* sequential reads at this point. It's skipping back and forth across 4 tracks (2 * stereo). This is almost becoming "random reads", though with any luck the head won't have to seek very far. Random reads normally cuts your sequential read speed in half, unless you use really large clusters. Large clusters can become a drain on drive cache and system resources, but it shouldn't be too big a deal if you have gobs of RAM to buffer the data and a fat 8MB cache on the drive to help out.
The fact we're basically doing random reads cuts our sustained read speed down. Not a ton, yet, since we aren't tracking a bunch of stuff yet, but noticeably. A good back-of-the-envelope figure to work with is that your minimum sustained random read speed is 1/2 of your maximum sustained sequential read speed. That's not entirely accurate -- smaller platters and faster motors reduce seek times and can increase this -- but it's a useful guesstimate on maximum throughput.
The problem is, writing to disk subtracts its slice from this random read time. Your writes, however, will almost always be sequential. There's only the single "arm" inside the unit, so effectively we're working with the same statistic at this point for writing to disk while reading large numbers of tracks: roughly 15 megabytes per second in the "pool".
Well, we did the math, right? Each stereo track sucks up a little over half a megabyte per second at 24/96. Therefore, we can record or play back roughly 30 tracks simultaneously, in any combination, before running out of bandwidth.
This is, of course, assuming a single disk. If you add a mirror to it, you can multiply the number of tracks you can run by roughly 0.86 -- or nearly double.
Some disks are faster than others. Seagate 15,000 RPM drives, for instance, can routinely pump out over 50 megabytes per second sustained sequential reads.
These back-of-the-envelope calculations, unfortunately, go out the window when you're not using a dedicated audio disk. If you are sharing with your system disk, your throughput will be dramatically lower. Often, on a heavily-fragmented volume with a large paging file, tremendously lower. Like 10-16 tracks of playback, rather than 30-something.
Of course, you can compensate for this. A lot of people just go with 16-bit, 48KHz recording on their laptops for just this reason. Look at the difference in bitrate:
24 bit * 96,000 samples per second = 2,304,000 bits per second
16 bit * 48,000 samples per second = 768,000 bits per second
2,304,000 / 768,000 = 3
By downgrading your sample rate, you've just tripled your recording capacity.
The question is, what do the extra 8 bits and 48,000 samples per second buy you?
1 word: Headroom (well, maybe it's two words, but I use it as one).
Your average listener can't tell 48KHz from 96KHz. They can't tell 16 bit from 24. An experienced listener will tell you that a 16-bit, 44.1KHz recording will sound "metallic" or "cold". What they are hearing is "aliasing" in the recording. A subtle thing. At 48KHz, 16-bit, on a pristine, dry recording of an instrument, only the most expert of sound engineers can hear the difference between that and a 96KHz, 24-bit recording.
But once you start adding effects, aliasing becomes more pronounced. The effects have to make "best guesses" at how to average between samples in order to create a smooth sound. The higher sample and bit rates give you the headroom to add those effects with a minimum of aliasing.
The bottom line is, if all you want to do is record 16 tracks of your band playing live, and you're not going to do a whole lot of tweaking afterwards, 16-bit, 48KHz will be fine and will give you a lot of bandwidth to spare on an internal 7200 RPM drive. If you're like me, though, and frequently run 30-50 tracks of audio in a studio, that internal drive will require an awful lot of mixing down and "archiving" unused tracks in order to do the job.
Don't get me wrong, you can do the job just fine. I did a ton with old 4 and 6-track recording gear back in the day, and you can make some amazing music. You just make it a little more slowly, with a few more technical hurdles to master before you can pump out astounding stuff.
Good luck; it's a fun adventure. I've been recording in home studios for 16 years, and it's always a blast.