[RichAplin]

Hey joy, I'm an idiot.
My math was wrong (I thought it was odd) and the 64k RAM will be just fine (at least for normal density disks);
Ok so the only legal combinations of bits off a disk are;
"[1]01" - 2 x 2us = 4us interval
"[1]001" 6us interval
"[1]0001" 8us interval
And this is what we see in the video I put up; three distinct peaks, corresponding to those combinations.

Mkay so if the track is 200,000us long, and our bits as shown above are 2us apart, we have 100,000 bits. Let's say 110k to account for long tracks.
If every bit pair is "[1]01" we'll have one transition per two bits, so 55k transitions, so 55kbytes. That's the absolutely worst case (and of course a track like that compresses well)
Other cases get progressively better of course (1:3, 1:4 etc). So, the RAM thing should be no problem with 64k, (we'd only have to get jiggy if we're reading/writing high density disks, and even then only if we're in 'raw' mode (i.e. sending one byte per transition))

Yay!

Eclipse: Yes we would, but we can do the buffering on the PC side where life is easy.

Does anybody care about amiga HD disks? I doubt anyone did any copy protection on them anyway(!), and we'll be able to read+write known-format ones fine. (with a known format you just decode to MFM bits instead of storing each bit timing, so less data, a track will fit in RAM)