It's also very much worth checking the "K2600 V2 supplement" manual here: http://www.kurzweilmusicsystems.com/Pro ... =downloads
...and reading Chapter 2 about "Triple Modular Processing" as this was a significant (free) operating system upgrade that took the K2600 from being a rich-mans K2500 and gave it its own unique functionality (i.e. instead of having just *three* DSP blocks it allowed up to *eleven* to be chained with various routing options), this pretty much blitzed any restrictions that the K2000/K2500 had (one noticable issue being the lack of a 2-pole high-pass DSP block meaning that you couldn't have a variable resonant high-pass filter in the chain after a DSP generated oscillator).
The thing about the Kurzweils (which, looking at my kit list you'll know that I'm biased towards) is that there are often ways to achieve something that there may not be a dedicated function for (e.g ring modulation, FM etc).
As a single "workstation" they really cover a lot of bases (especially considering their age) including sampling (the sample option literally just gives you the physical sample inputs but you can import .wav and .aif samples via floppy/SCSI and use these on a K2000/2500/2600 which doesn't have the sample option installed), Virtual Analogue (each DSP block can be a real-time DSP-generated oscillator), ROMpler (built-in ROM samples plus optional Piano/Classical/Contemporary ROM expansion boards), multi FX (with KDFX - an option on the K2500 and fitted as standard in the K2600) and a very flexible sequencer (not that I use it much but it's arguably on a par with that in an MPC).
A few thing about the Kurzweil voices (good and bad):
Each layer uses 1 voice (if mono) or 2 voices (if stereo) therefore a stereo sample uses up two voices (one for the left channel, one for the right channel). A 32-layer "Drum Channel" therefore uses 32 voices (if all layers are used and sounding at the same time.... even more voices if the layers are stereo).
A Triple Modular layer uses three voices (as it's effectively three single voice layers chained end to end)....
however.... a normal program (i.e. non-Triple Modular program) can consist of one ROM/RAM sample plus three DSP generated oscillators per layer - this only takes up a single voice. So effectively a 48-voice K2500 can actually play 192 oscillators simultaneously (48 being ROM/RAM samples and the remaining 144 being DSP generated).