By no means do I consider myself an expert, so take my opinion for what it's worth ...
The first thing I'd ask myself here is "if I'm using a /8 of RFC 1918 space, why be so conservative?". Given your requirements, unless the client goes through some seriously massive growth you have an *enormous* pool of address space to spare that can never leave the confines of their network. I'm not advocating being blatantly wasteful with the space, but simplifying the design will make management easier in the long run.
I'd consider allocating a /22, or even a /21 per site from day one and just keep it simple. Each office gets broken in a standard way ... i.e. first two /24's for employee access, last /24 for infrastructure, and whatever else in between. I'd also pick a random base point to reduce the likelihood off colliding with any other networks that I may need to interconnect later. Plus, if there's already a couple spare /24's per office, I won't need to modify the WAN's routing tables if I need to light them up and use them (not that this is a big deal, but the WAN routing always stays simple too).
I've inherited networks that were carved into pieces as small as /29's and /28's using 1918 space and I could never understand why those before me would artificially constrain themselves ... not to mention cursing them as I try to expand things (renumbering is a PITA no matter how small the block).
I save the brain power for efficient design when I'm dealing with direct allocations where every address needs to count ... in this case, I'd probably lean towards simplicity and scalability vs efficiency. But, by all means, feel free to be efficient if that's what you prefer.
