The audio engine, fundamentally, is typically the same for both EAX-enhanced and non-EAX platforms. In some cases, developers will utilize EAX extensions to achieve certain results that may be typically processed in software (such as distance attenuation), but there's always a non-EAX software fallback. The audio engine is doing things differently by making EAX calls, but it's not a different engine, nor are audio engines somehow inhibited by non-EAX hardware.
And, actually, the inverse of your last statement is true. Without EAX, developers would rely more upon the processing possibilities of APIs and software engines. With OpenAL, you're given the capability to do API-wide effects and are given a great deal of control over how they function. The limitation is the number of effects sends. If you aren't using OpenAL, you can always build these effects by hand via the audio engine. The audio engine becomes more complex and more robust out of necessity. Crysis is an example of this.
For games that support EAX, they're just using EAX extensions and having the engine pass the variables. They can fine-tune the variables that get passed, but they're also potentially limited in how extensible these effects can be. They can also only utilize these extensions if the user has some sort of EAX-capable hardware.
I think EAX is an excellent tool, but, like 3dfx's old Glide API, later versions are a proprietary and hardware-specific means of achieving the end result. EAX is a good thing, but it's also a very bad thing. Those that fail to see this duality are somewhat blind to reality in my opinion.
Except I wasn't talking about EAX.
Developers will not create a sound engine that eats up 30% of a CPU. They will develop a sound engine that will eat up the smallest amount of CPU possible (within reason). If they know a decent amount of consumers have hardware accelerated sound cards (regardless of EAX or other reverb effects, just has DS3D/OpenAL hardware accelerated), they will develop a more immerse audio engine (because they won't be afraid of a massive performance drop). If we dropped hardware acceleration altogether, there will be little reason for developers to improve and make their audio engines more complex than it is today (and trust me there is a lot of room in this area).