Stephan is entirely correct: the OA APCs Elsevier charge are not directly dependent on whether or not you are a subscriber. The method used by Elsevier, and most publishers, is to reduce the overall subscription cost in proportion to the number of OA articles in a journal, not reduce the APCs in proportion to subscription spending. In theory, were a hybrid journal to go 50% OA, then the subscription price would be halved - which makes sense - but as David notes, the mechanics of this are pretty opaque. Most journals are not bought at list price, and things like varying numbers of articles year-on-year confuse things.
That said, there are projects which run the way you envisage it working - where APCs are reduced or waived for subscribers. The recent UK initiative to put substantial government funds into gold OA has led to a lot of pressure to come up with a system like this, because otherwise we would create a situation where UK universities spent very heavily on gold OA but only saw small reductions in the global subscription rates. Similar systems are being set up in Austria and elsewhere, and I suspect will become common over the next few years.
At the moment, two of the major publishers (Taylor & Francis and Wiley) offer such "offsetting" schemes. The T&F one is summarised here and the Wiley one here. With these, the subscribing institution gets a "rebate" voucher based on their subscription (or subscription + APC) spend over the previous year.
They can then use this to pay for future APCs. Elsevier do not have such a system, but JISC are strongly pushing major hybrid publishers to set them up, and I feel it's likely we'll see one for Elsevier within the next couple of years.
Edit: the RCUK Open Access Review, released today (26/3), has a little more detail (Appendix H) on the current offsetting projects - as well as T&F and Wiley, there are active projects from the Institute of Physics, the Royal Society of Chemistry, and Sage. The Sage one works exactly as the original post assumed - authors at subscribing institutions are charged a heavily discounted APC - while the others are a rebate/voucher system based on total spend.
Edit2: and as of 31/3, Springer have announced a very vague offsetting plan with JISC. No details on how it works yet but probably some kind of voucher system.