I have brought up my issues with Core Sound in general and their Mic2496 in particular several times on these forums. I don't feel the need to go over this again, suffice it to say I'm just not impressed with either, and I would urge caution with anyone considering dealing with coresound.
However, I will talk a little about the technical ins and outs of recording at 24/96. I think I can speak with some authority on this, since I did my research 2 months ago. One side project I'm working on (VERY loosely working on, this is strictly a hobby, paying work must come first) is a stealth recording device. The main design issues are (1) battery life (2) recording capacity and (3) write speed to the media. Your choices are basically either compact flash or some form of hard drive (e.g. microdrive). The first two should be fairly obvious: CF has much better battery life, principally because there is no motor to spin and no heads to move over the platter; however current state-of-the-art CF is I think 5GB, compared to several hundred GB for a conventional hard drive or I think 60GB for the 1.5" Toshiba drive used in an iPod. The math is easy: 24/96 takes just over 0.5MB per second of recording time (WAV), so a 4GB CF card would hold around 2 hours of recording, compared to the 30 hours or so for a 1.5" 60GB drive.
The 3rd item, media write speed speed is where things get interesting. As I noted above, 24/96 requires just over 0.5MB per second. I took a look at the specs for SanDisk CF cards, they are rated at 16MB/s burst transfer, which should comfortably enough, right? Well, not quite. In order to get this rate there are some base requirements. You need to start writing on a page boundary (typically, on multiples of 8 sectors), and you need to write a power-of-2 (2,4,8,16...) number of sectors in one go. The more you write in one burst, the better throughput you will get, throughput tends to peak around the 128-sector mark. And here's the crunch: if you don't meet these requirements, throughput drops
significantly. That's why you need to pre-allocate the file, to get as close to this ideal as possible (also, so that you're not seeking all over the media, between writing audio data and writing updates to the FAT). But even so, the underlying operating system (PalmOS, WinCE) will generally not handle things optimally. Microdrives tend to suffer the same problems; external drives much less so, by virtue of having large on-drive buffers.
It CAN be done, with careful design, and possibly writing your own realtime operating system dedicated to this task, there is plenty of slack in a CF/microdrive system to make it work. The solution is not to throw more CPU power at the problem, since the CPU is not the bottleneck. Careful design is the correct solution
Embedded systems is where I'm at, it's what I do for a living. This would be an incredibly cool project... if only I could find the time
best regards,
stephen