I'm sorry to deviate this thread slightly to the digital side of things, but I have some experience in the field, and would like to offer some insight, and at the same time recap (pun intended) the thread this far. The whole thing is written with the "tape as an effect" mentality. Sorry if you don't like it.
1. I/O electronics.
This is the least important bit in the overall sound, but it order to drive those heads a basic gain and attenuation is needed. This is the easiest thing for us to replicate as there's a choice of multitudes of good line amps. In digital form this type of gain staging hasn't been emulated as of yet, and that's what a soon to be released plugin of mine will do. I've christened the effect as intra-modulation (IM) and it somewhat captures the interaction between transformers and gain stages of electronics, and several other secondary effects that go nicely "wrong" in any analogue circuit.
2. pre-emphasis/de-emphasis.
Exteremely important factor in the saturation performance and should not be omitted. Also easy to replicate in either digital or analog form. What's your favourite emphasis curves? Simply DIY that or similarly do a digital oversampled filter using the same curves. This is the only place where impulses of the real units would do any good (and even then only if sampled at both stages, which would be difficult). Dynamic convolution is only good to recapture the filtering performance, and it might as well be done algorithmically instead (cheaper DSP wise, and easier). This was done in the voxengo TapeBus mentioned earlier in this thread, and it features, for example, Studer B67 impulses by yours truly. Its sound still leaves a lot to be desired.
3. Magnetisation.
There's very little info on how to do this digitally, and I'm doing research on the subject as we speak. Electronically one might as well do the portico thing and use real tape heads if one is willing to gut working units. If any of you manages to do this, I would be very interested in any kind of analytical measurements on it. Basically one could emulate this as a waveshaper in digital form, but the actual response curves and shapes are unavailable in a pure magnetisation form. The saturation dynamic curves are always expressed in conjunction with the tape media curves, unfortunately. Maybe one of you would have data already available?
4. Tape media.
Generally when tapes are emulated magnetisation and tape media characteristics are filed under "saturation". This is a mistake, since both play nearly equal role in tape sound. Tape media is most likely impossible to emulate electronically unless using tape loops, but that creates the many problems already covered earlier in this thread. Digital implementation is another thing. I'm doing research on this subject as well.
Basically the media has little influence on the sound when it's not driven (well, it does a little actually). Things change drastically when the tape is stressed, and we start to hear all the good sounding fluctuations that now get exaggerated. Using a separate digital waveshaper with time-based randomisation chanracteristics should work. The model should be dynamic, ie high saturation exaggerates the "smear" while quiet sounds get very little of it.. Problems arise when determining the "right" fluctuations as it varies from tape manufacturer to another. Data on this is mostly unavailable, as it's time based and difficult to measure.
5. (concerning only digital implementation)
Analogue electronics produce lots of secondary side effects: thermal coupling, physical limits of transistors/diodes etc. While it isn't anything to worry about when DIYing a tape effect, it has to be modelled separately in a digital form. I already mentioned IM, and since these effects are minor in the tape sound, IM is perfectly adequate to model them.
What did I miss?
*passes the baton*