Transistors have a gain that drops with Ic. If you have no load, you don't see that drop. With load, you see a lower gain at the frequency you are looking at. That manifest itself as a faster roll off.
If you are simulating using ac analysis then the response is linear about some initial operating point, thus the effects like changing beta (with Ic) and voltage level shouldnt show up.
But as Bonsai points out the R/L is different for each run ( 2uH / 4 ohms is pretty close to 200Khz, but there must be some additional contributions).
You should include an input filter and I always recommend an output inductor /Zobel network.
Try running your sims without the the input filter and output inductor/Zobel network with a capacitive load to get a measure of how load tolerant your design is. Start at about 100pF and work your way up to say 2uF in half decade steps. You can use the .step param command.
high Q uF is unreasonable load for a audio amplifier - not even seen driving ESL step up tansformer primary where the winding series R and leakage L decouple the panel C above ~50 kHz
cable C for most domestic setups, conventional cable will be < 50 pF/ft
with Spice you should be able to measure with filtering/Zobel in place - just measure the right spots, remember you can do real world impractical calculations, normalization with most Spice' viewer waveform arithmetic
" . . . high Q uF is unreasonable load for a audio amplifier."
Precisely. That's why you need to do it to see the limits of the amp.
I agree that the cable inductance will help to isolate the capacitive load, but I see too many designs where people chase low THD at the expense of stability margin, which in fact is a significantly more important parameter.
I dissagree with the bi or tri-amp situation for home HI-FI. That should be more suited to a vehicle where listening space is limited. It is too common now with X.x stereo models with all the satilites and seperate sub. Cheap amps and low power speakers = more profit! I think the speaker(s) should be a single full range load on the amplifier. The amplifier should be able to drive that load without issues. The less processing of the signal the better IMO. Unfourtunately most amplifer circuits are not capable in one way or another of doing it without some sort of coloration. It is part of why there are so many people out there making thier own designs.