My understanding is that the difficulty in producing truly random numbers that could be used for one-time pads has more to do with the rate at which they are produced than the quality of the randomness. I believe sources of entropy ̶y̶o̶u̶ ̶d̶e̶s̶c̶r̶i̶b̶e̶ like interrupts from the disk drive (in addition to sources like interrupts from keyboard inputs) are typically used to provide seed values for popular cryptographically-secure pseudorandom number generators (CSPRNGs) like Fortuna and Yarrow. As long as statistical tests for randomness are applied to the data from these sources of randomness, and attackers are not capable of modifying inputs to reduce their randomness, these sources of entropy are indeed quite good.

But if one-time pads were to be widely implemented, it would be necessary to constantly generate very large quantities of random numbers to meet the requirement that the keys be at least as large as the plaintexts. As far as I know, the rate at which we can currently generate truly random numbers could not easily support the huge amounts of encrypted data transmitted in today’s world if we were to attempt to replace algorithms like RSA with one-time pads. (Obviously, other limitations like secure key distribution would make replacing an asymmetric cryptosystem like RSA with one-time pads a quixotic task.)

I would be interested to learn about any approaches to quickly generating high-quality entropy on consumer devices if there are in fact ways to do this at a rate that could support more widespread use of one-time pads in contexts that involve large volumes of data.

(response edited with strike-through to reflect we are talking about different sources of randomness from the disk drive)