The question posed came about during a 2nd Year Comp Science lecture while discussing the impossibility of generating numbers in a deterministic computational device.
Th
The approach of measuring something to generate a random seed appears to be a pretty good one. The O'Reilly book Practical Unix and Internet Security gives a few similar additional methods of determining a random seed, such as asking the user to type a few keystrokes, and then measuring the time between keystrokes. (The book notes that this technique is used by PGP as a source of its randomness.)
I wonder if the current temperature of a system's CPU (measured out to many decimal places) could be a viable component of a random seed. This approach would have the advantage of not needing to access the network (so the random generator wouldn't become unavailable when the network connection goes down).
However, it's probably not likely that a CPU's internal sensor could accurately measure the CPU temperature out to enough decimal places to make the value truly viable as a random number seed; at least, not with "commodity-class hardware," as mentioned in the question!