After a bit more digging in the forums, there is quite a bit of chat about different clock frequencies and because of the way that the millis and micro delays are generated, its basically 8M or 16M for them to be meaningful. If you use different xtal frequencies, you still get a delay, but it wont be what you are expecting. This seems to be pretty much hard coded into the core files with other functions and libraries making the assumption of 8M or 16M xtal
I used a ferrite bead because of some comments on the Goldilocks Google design docs page. I thought I'd give it a go. I have always used basic inductors so far but I'm willing to try and see. In my layout it may not make any difference as its hardly set out for optimum analog performance. On the plus side, it uses linear regulators so that will probably help.
As for the crystal, here is an example of the package http://au.element14.com/abracon/abm8g-16-000mhz-b4-t/crystal-16mhz-18pf-smd/dp/2308711?Ntt=2308711
This is just to show the style. Its not the friendliest thing to hand solder. I'll have to get a toaster-oven reflow setup going.
Two pins are the xtal, the other 2 are Gnd.
I use the 10u caps because I have a stack of them. In the Wiznet forums, good decoupling on the wiznet module seems to be fairly critical. I also wanted to keep to a 1206 footprint and I did not want the package height to be too much, so multiple smaller devices.
The Wiznet module has a built-in pullup resistor. I put the others on because while using the ISP programming, the micro I/O pins float so the CS lines could become active and then random data may appear on the MISO line and cause the programming to fail. I'd just rather err on the safer side as I dont like having to tack bits on later.