If, per say, you did have 1GB of cache, it would be significantly faster than having 1GB of system memory, at least after the first fetch. There is a travel time along a line, plus propagation delays through the memory controller into cache, which depending on how loaded the system is, could be anywhere from 10s of CPU cycles to hundreds of CPU cycles, since the bus ticks, typically, once for every 5-15 (approximately, depending on the setup) CPU cycles. A memory fetch may take several ticks of the bus, plus in memory lookup time, before the cache receives the requested memory block.

However, since having that much cache is probably quite a ways off, having a good amount of system memory with a good bus to prevent having to hit the hard drive, is about as good as you can do. Give or take a few small bonuses (multi-pumped buses, dual-channel, etc...)

Simplified:

DRAM (regular system RAM) is very fast after the first fetch. It's opening the page that's the real time waster.

Cache is "instant" because it doesn't need separate page addressing like DRAM (because it can directly drive the data lines).

You could probably build 1 GB of SRAM, and it would be very fast (as it would be low latency), but it would use lots of chip area (much bigger than a 1G stick of DRAM) and, more importantly, it would use lots of power (and thus need cooling).

It would also cost a fair bit :-)