Using Jenkins 1.495.

I have a standard diamond-shaped job configuration. A starts B and C in parallel, and if they both succeed, starts D (using the join plugin).

At seemingly random times, D is not being run, even though B and C are both successful.
When I look at the console output of B, I see:
. Notifying upstream build A #80 of job completion
. Project A still waiting for [C] builds to complete
The console output of C is the same, vice versa:
. Notifying upstream build A #80 of job completion
. Project A still waiting for [B] builds to complete

... so, no sign of D running. Most of the time, however, the jobs work fine.

Some additional info that may or may not be helpful:

• A, B and C are matrix jobs. D is not.
• D is run from A as a parameterized job (parameters are explicitly set).
• I actually have a lot of different A, B and C jobs, but only one D job, which is being called with different parameters. The D job has 'concurrent builds' enabled. The time when I observe the problem seems to match the time of multiple concurrent builds.
• This shouldn't be a race condition, as the B job takes about 30 minutes and the C job about an hour.
• I have some slave nodes connected, so the jobs are being run on different nodes. However, all matrix jobs (A, B and C) have their parent job 'tied' to the Master.

Can you tell me what other information I can provide, or what experiments I can carry out to isolate the problem?