I doubt rollers can do ramp tests reliably, the idea is that the smart trainer increases resistance in steps until you absolutely cannot push the pedals any more, then take approx. 75% of that max power as your FTP. So you need something that can increase resistance consistently, and at the same time hit a point where the resistance is too hard for you to push the pedals.
Unfortunately ramp tests can be inaccurate too. The standard Zwift ramp test starts at 100W, then increases by +20W every minute. The power ramps up too fast, and too steeply - for people with very good anaerobic power (or "speedsters", as @
colnago62 alludes to), a lot of that initial burst of power contributes to the test, allowing them to quickly hit high amounts of power with this test (and therefore overestimates their FTP value).
Zwift has introduced an alternative ramp test for "light riders", though I have not seen or tried that one to see how that works.
TrainnerRoad's ramp test, from online reading, adjusts the starting watts and the ramp rate according to a percentage of the rider's current FTP, so supposedly is more accurate.
The Sufferfest's ramp test, called the Half Monty, also adjusts its baseline based on your FTP and, if I remember correctly, has a longer duration per step. In addition to that, there is a 20-minute interval section after the ramp test which is not as hard as a full FTP test, but feels more like a tempo test, to fine-tune the FTP and lactate heart rate obtained from the ramp test.
But ramp tests aside, some FTP test protocols start off with a 5-minute hard interval effort after warm-up specifically to drain out your anaerobic reserves, before doing the 20-minute FTP test proper. This supposedly ensures that you are fully relying on your aerobic capacity for the test, rather than having that extra boost of "speedster" power tainting the FTP test. Not sure if this principle can be applied to the shorter 8-minute test though.
However, I don't think a standard FTP test will resolve the issue being discussed here and a ramp test would be more appropriate. The FTP test can be easily fooled by pacing issues and any other variability going on (and going by the discussion here, I believe the issue to be @
Carbonfiberboy's FTP being underestimated, rather than overestimated?). A good ramp test would completely eliminate those issues altogether, just keep turning those pedals until you absolutely can't turn them anymore.
Edit: One last edit. One (perhaps viable) alternative to FTP testing, is having a Garmin watch or Edge cycling computer auto-calculate for you. You need a power meter and a chest strap heart rate for this, and just go out and do your usual (hard) rides. I don't know how many rides it takes before it comes up with an estimate, but the values it has given me is pretty close to what I get when testing FTP.