"would it be better just to record @ 16-bit in the first place?"
Very rarely. About the only time I can think of that recording at 44.1/16 bit would be best, is if you are recording acoustic music "live to two-track" and intend to only release it on CD. Then, if the only digital processing you will do is to clip off silence - as in no eq, no reverb, no level changes, absolutely NOTHING done to the original files - THEN, recording to 16/44 would make the most sense and likely give you the best sounding CD. How good this would sound is also a function of hardware quality - better converters, etc, would give better finished product assuming the recording was done with all the best methods possible. (This is a truism often forgotten in the quest for "magic bullets" that will allow us to be slobs and still sound good)
However, the minute you adjust a level, add/subtract EQ, add reverb, chorus, flange, ANYTHING at all in the digital domain, you have just performed an arithmetic operation. When you take one 16 bit digital word and add it to another, you now have increased the bit depth of the information. Sticking to the use of only 16 bits, you've just decided to throw away some bits of information. Let's say you EQ a track at 5k, then chorus it to fatten it up, then add some reverb to push it back behind another track, and you do this to most of the tracks in your mix, all within your DAW - if you throw away all but the 16 most significant bits each time you change the track, the cumulative errors will eventually add up to where your song sounds worse than a 4th generation cassette overdub.
Now, taking that same mix, suppose you save 32 bits of information internally, plus add floating point math so you can be even more accurate with the calculations, and then ONLY ONCE at the end mix down and burn a CD, dithering down to 16 bit just for the CD - All during the previous operations, you have saved those "extra" bits of information and used them to calculate the next step, and when you finally DO throw away all but 16 bits, you've minimised the math errors by so much that there is a large difference in the percieved (and actual) quality of the final product.
As far as 16 versus 24 bit, one of the main reasons why 24 is better is headroom. In order to get anywhere close to true 16 bit quality on a 16 bit CD, you need to record at 0 dBfs, or within 1-2 dB of it. This means either a limiter to prevent "overs", and/or extremely anal gain riding or re-recording until you get it right. With 24 bit recording, you can lower the max record level by 6-8 dB, worry very little about "overs", and still record at least 20 bits of useful information. Then, during the course of tracking and mixing, you can optimise levels at 24 bit or better, and dither down just as you go to CD, to a CD-compatible 16-bit 44.1 kHz data stream.
Most of the better DAW software (Nuendo, Cubase, Logic, Samplitude, Sonar, etc) allows you to choose to work and save in 32-bit floating point, up to the sample rate your hardware supports - This allows you to only dither down and downsample ONE TIME when burning a CD, except for monitoring during tracking. This whole procedure minimises math errors and maximises sound quality of multitrack, multi-effected projects so you can get the best sounding final mix possible.
Case in point: Which do you think would sound better, a stereo cassette that was recorded from a mixdown of 6 synchronized cassette 4-track portastudios (same audio quality as a stereo cassette) or a stereo cassette that was recorded from a 24-track 2" tape machine ? Hint: with cassette 4-trackers, you're dealing with a tape speed of 1-7/8" or 3-3/4" per second, and a track width of less than 1/32", or .031" - the 24-track 2" is at least 15" per second with a track width of .083" - Sooo, you have 2.67 times the track width and 4 times the tape speed. What I'm trying to say here is that if you do a project at the same specs as the final product will be, then you lose quality that can be saved by doing all but the last step at higher quality.
The above is by no means a complete answer, but I hope it shows some of the reasons why it takes more than 16/44 to accomplish a final product that is even CLOSE to 16/44 in reality. Try to remember that so-called "CD quality" sound cards for computers generally aren't. The ad-speak version of "CD quality" really only means that the card samples 16 bits 44,100 times a second. It never says that 4 bits of that info is pure noise from putting a low level audio signal into a box full of electrical noise BEFORE it's digitized - Just one example.
The reason really good CD's sound that way is that no one involved thought "good enough" really was, anywhere along the way... Steve
