The K8 L1 architecture was inherited from the K7. It was originally designed for a K7 born when only one level of on-chip cache was possible.
AMD doesn't have the manufacturing capacity or design experience to go for monster add on L3s like Intel uses in its x86 line. To maximize performance with reasonable area constraints AMD should really redesign a three level cache hiearchy from scratch. L1 design is tightly integrated within the CPU layout and pipeline design of any processor. So it doesn't make sense for AMD to spend the effort at redoing the cache hierarchy until it redesigns the CPU core itself.