It highly depends on what software you're using to manage the keys. For example, the proprietary crypto library we use here splits the key over several memory locations, flips them big-endian / little-endian, etc and only reassembles them when they're needed. Any good crypto / key management tool will have its own bag of similar obfuscation tricks. This will make it hard for you to locate keys in memory unless you know how they've been obfuscated.
As you say in your question, TPM or full memory encryption are probably better uses of your time. Unless you know exactly which crypto libraries / key management tools are running and are willing to reverse engineer them / add key destruction hooks into their source code.
Update: as you say in comments, security through obfuscation is frowned upon as the only source of security (ie it doesn't protect you from highly motivated attackers), but it does still have its uses.
Update #2: having chatted with a colleague about this, it sounds like you're trying to make your server into an HSM, these are hardware devices specifically designed to manage private keys, and to wipe them if any kind of tampering is detected. If your organization is concerned about attackers having enough physical access to pull off a cold-boot attack, then you should really be using commercially-certified HSMs and not have private keys in memory at all.