Many techniques required for the first are useless for the second: if everything is open source, or even if it's not and the attacker is smart enough (unlike most protection measures which are supposed to be independent of intelligence...), then measures implemented by this tool like 'virtualization' (i.e. indirecting execution through an interpreter), branch obfuscation, etc. can be undone easily by an attacker with full access to memory; if the attacker doesn't have full access, there are much simpler and more performant types of randomization that are equally good at defeating them. And measures useful for the second (generally, anything that involves multiple variations of the same version of a program, to force exploiters to learn which one is in use) are no good against the first (since the attacker can just pick one to extract secrets from).
As an example, in Adobe's RTMPE (encrypted RTMP) protocol, the type 8 encrypted handshake used an obfuscated crypto routine that comprised 95KB of x86 object code, taking billions of machine cycles to execute. After de-obfuscation it was found to simply be XTEA, which is implemented now in rtmpdump in only 4 lines of C code and executes in hundreds of machine cycles.