You CAN use USB risers in folding, but higher-end cards suffer a significant performance hit if you do so, ballpark 25-30% in some brief testing I've done on my Aorus GTX 1080 ti cards.
If you were seeing 500k on one, you either had a "pain" work unit type or there was something else limiting the card, even on a USB riser I normally see more like 800-900k on most work units (vs 1.2M AND UP on most work units on non-riser setups).
The hit is less on lower end cards, but still noticeable.
I've been folding on and off 10 years. CPU with a AM64, than waaaay back on a 9600 GSO (old folders remember this being a beast of a card for performance ratio). I've done testing with many cards because I did find a fold forum post on folding forums that someone testing this waay back in 2010 that the OP said he/she found no difference. This was true at the time because the cards were not powerful enough to saturate the full bandwidth on even 4x PICE 2.0. Things have changed a lot in the last few years. I'm looking at my 1070Tis right now and they using 56% bandwidth of PCIE 3.0 8X or roughly 4.421 GB/s if you do the math on it. PCIE 3.0 1x is 984.6 MB/s. That means a single 1070Ti is using 4.49 times the bandwidth of 1x 3.0. Now that is not saying you will lose 75% when on USB risers. Its project dependent but with 1x I was around 330K and with a 8x slot (not fully used) it is reporting 768K PPD on average. Same card, same computer, same project.
Edit: Also don't forget folding really does need a CPU core per card or it will suffer more than a USB riser would do to it. Test it out. Put you CPU to full and watch you card PPD drop to nothing.
Food for thought....
Edit 2: Oh I also forget that any USB cannot exceed 500MB/s. Thunderbolt can, USB cannot, so really even if its reporting being in a 3.0 Slot, its more like 1x 2.0.