I now get about 4 KiB/sec on large streams (such as 64 KiB) reads from srandom(4), with 8 KiB/sec initially, in contrast to the less-than-100 Bytes per second (300 Bytes per second initially) without the eKey.
Of course, there’s still room for improvement – I fixed the ioctl(2) calls, removed strcpy(3) and sprintf(3) calls, and added arc4random_buf(3) calls for generating the nonce (which can now be made much larger than the 12 conservative bytes the original code reads from urandom(4)), and made it work at all on our platform (and, possible, OpenBSD). But I get statistics now, even if told that my ekeyrng mksh(1) script is “Cute”.
Make sure you update to at least luasocket-2.0.1-1 for some bugfix (pkg_info(1) has a bug preventing it from seeing that – what worries me even more are some outputs not sent with the mail), and that you have a recent kernel (post the “wtf ist hallowe’en” snapshot!) since lsusb (even when ported) doesn’t output anything, and nobody knows what arguments to ekey-ulusbd are needed to make it find the eKey.
ObInfo: new CA bundles are out too, and more binary packages.