I just committed a change that should fix the by-now well-known
bug with 3c590 (and 3c900?) PCI devices hanging under load.
The fix  should show up in tomorrow's sup.

I haven't ever managed to replicate the exact problem reported,
(e.g., ping failing) but I've duplicated something similar on a
100mbit net and verified that  the patch indeed fixes the problem
I see here.

I haven't been able to verify that first-generation Elink-III
devices (e.g., 3c509, or the PCMCIA or EISA equivalents) still work.
If anyone has problems with them, send a PR and please also let me know.