RNA networks, a three-year-old Oregon start-up, claims to have done what nobody else has done before and make it possible to share server memory.

It calls its breakthrough “memory virtualization” and says it’s the third wave in virtualization after server and storage but technically speaking it’s not really virtualizing the memory – it’s not even designed to work in virtualized environments – it’s just making it possible to share all the memory in the data center across the data center on-demand and, virtualization or not, that’s still a breakthrough.

Memory is the most constrained item in a system and if you don’t have enough, well, you have to swap data in and out and the system’s ass can drag.

Remedies so far have included scale up and out, server virtualization, faster storage, in-memory databases, data grids, I/O virtualization, solid state drives and storage caches.

Now little RNA comes along with a newfangled Memory Virtualization Platform that promises to wring high-performance computing from existing commodity hardware by decoupling memory from the processor and server and turning the memory into an application-aware shared network resource available to whatever physical servers need it.

It fancies its software is the killer app of the enterprise data center and claims first mover advantage in leading data center transformation into the next decade.

It says it can save users money by way of server consolidation, smaller physical footprints, reduced over-provisioning, deferred server upgrades and increased utilization.

It says to model a 37% reduction in network costs and 42% power savings.

RNA’s widgetry pools or aggregates the available memory across nodes and makes it available to all servers in the data center. Servers can either access the pool, contribute to it, or both. Put another way, it creates a clustered cache that interconnects the memory across all compute nodes and will dynamically scale. There’s no limit on the number of nodes.

The start-up’s first product based on the platform is a thing called RNAmessenger aimed at high-volume low-latency transactions like at hedge funds – specifically aimed at financials actually – at least the ones that still have transactions – but can be applied to any business-critical environment, like online gaming and content delivery, demanding superior transaction processing speeds.

RNAmessenger, which just emerged from beta and only works on Linux, claims better than a jaw-dropping 500% improvement in trade execution latency. “By keeping all trade execution messaging in memory,” it says, “hops are eliminated and unmatched latency (microseconds) can be achieved.”

It reportedly works on any size message.

It costs $7,500-$10,000 per node for a perpetual license. Support is extra. The memory is managed by a rack-mounted x86 server appliance with 16GB or 32GB of memory of its own.

RNA has flown pretty much under the radar until now. A year ago the 25-man outfit got a $7 million A round from Menlo Ventures, which backed 3PAR, F5 Networks and IronPort. Initial backers reportedly included Oregon Angel Fund, Divergent Ventures and Reference Capital.

So far RNA’s collected 10 customers, including an unnamed multi-billion-dollar global hedge fund that’s reportedly getting a staggering 840% improvement in transaction volume. RNA will be looking for OEMs.