I have a similar development setup and have made effective use of ocp-build to enable a tight development loop while still supporting opam deployment. Suppose you have libraries A and B and a program P that uses these, each in their own directory (or repository). You set these up with a per-project .ocp file that describes the library / program sources and a per-project Makefile. For each project X, the Makefile (or OPAM build rules) executes:
ocp-build init
ocp-build configure
ocp-build X
This configures ocp-build to only search in that directory when doing its build. When doing development, you skip the per-project makefiles and instead arrange things as:
projects/A
projects/B
projects/P
And you execute the "ocp-build init; ocp-build configure" portion in the *projects* directory. This sets "projects" as the ocp-build root directory and causes ocp-build to find all the per-project .ocp files. The dependencies between the program and libraries will be properly resolved and even if you have A or B installed with findlib, the local versions of the libraries will take precedence when building / linking P. Just compile P with "ocp-build P" (executed somewhere in the projects/ directory tree). You do have to be careful to ensure that during development you never accidentally create a "projects/X/_obuild" directory (e.g. you never accidentally invoke "make" in any of the project directories). ocp-build will stop at the first _obuild directory it finds, so having _obuild directories other than projects/_obuild will prevent the full traversal of the projects/* subdirectories.
--Stephen