From an anonymous author, a follow-up to the discussion about the cache etc.:
counterpoint 1: Itanium, which was EPIC like Elbrus, failed even with Intel behind it. And it added prefetching before the end. Source: https://en.wikipedia.org/wiki/Itanium#Itanium_9500_(Poulson):_2012
counterpoint 2: To get fast, Elbrus has also added at least one kind of prefetch (APB, "Array Prefetch Buffer") and has the multimegabyte cache that Zaitcev decries. Source: [kozhin2016, 10.1109/EnT.2016.027]
counterpoint 3: "According to Keith Diefendorff, in 1978 almost 15 years ahead of Western superscalar processors, Elbrus implemented a two-issue out-of-order processor with register renaming and speculative execution" https://www.theregister.co.uk/1999/06/07/intel_uses_russia_military_technologies/
1. Itanium, as I recall, suffered from the poor initial implementation too much. Remember that 1st implementation was designed in Intel, while the 2nd implementation was designed at HP. Intel's chip stunk on ice. By the time HP came along, AMD64 became a thing, and then it was over.
Would Itanium win over the AMD64 if it were better established, burned less power, and were faster, sooner? There's no telling. The compatibility is an important consideration, and the binary translation was very shaky back then, unless you count Crusoe.
2. It's quite true that modern Elbrus runs with a large cache. That is because cache is obviously beneficial. All this is about is to consider once again if better software control of caches, and their better architecture in general, would disrupt side-channel signalling and bring performance advantages.
By the way, people might not remember it now, but a large chunk of Opteron's performance derived from its excellent memory controller. It's a component of CPU that tended not to get noticed, but it's essential. Fortunately, the Rowhammer vulnerability drew some much-needed attention to it, as well as a possible role for software control there.
3. Well, Prof. Babayan's own outlook at Elbrus-2 and its superscalar, out-of-order core was, "As you can see, I tried this first, and found that VLIW was better", which is why Elbrus-3 disposed with all that stuff. Naturally, all that stuff came back when we started to find the limits of EPIC (nee VLIW), just like the cache did.