I'm quite confused by the question, but I'm going to take my best shot. Aside: Normally I strongly dislike this approach, but intuition tells me that nagging the OP to clarify questions that s/he admits s/he doesn't fully understand is a worse violation of SEC:SE principles.
RSA, and all public key encryption, is very processor intensive. I'm not surprised that you have performance problems with a strong RSA key. Symmetric key encryption is less processor intensive, but involves a fairly complex key management overhead, which is usually at least partially manual. Manual key management is painful.
ECDH is a hybrid solution. Perform encryption/decryption with a symmetric key (Good performance), and use the public keys to generate and manage session keys (solid key management with high automation, low overhead and low pain threshold). Best of both worlds; you get the performance of symmetric keys without the key management overhead.
That is the 50,000 meter theory; if SEC:SE were a technical discussion site we could delve into the details.
The second set of questions is much more problemmatic. Bruce Schneier has a pithy quote that I can't call to mind right now to the effect that 90% of the problems with crypto are not in the design, but in the implementation. How much performance improvement will you encounter? Depends on the implementation? How strong is the encryption? Depends on the implementation.