Unfortunately, parallelizing the build steps is not straightforward. Each step 
has to run in its own Docker environment. Therefore, we would have to setup an 
Ubuntu OS, all dependencies and the Fast Downward clone for each parallel step. 
This makes the pipelines script more complicated and doesn't reduce the time 
for testing. We could revisit this if we ever create Docker images with all 
Fast Downward dependencies. For now, I'll strip the commits from my clone and 
will close this issue.

Feel free to go ahead with this as you see fit. (No need to review, but I'd be
interested in seeing the diff once you're done, if there is something to
configure in our repository.)

They also added a "size: 2x" option (https://confluence.atlassian.com/bitbucket/configure-bitbucket-pipelines-yml-792298910.html#Configurebitbucket-pipelines.yml-ci_sizesize), but this 
seems to affect only memory (4 GB instead of 8 GB) and not the number of cores (8 in both variants).

Recently, Bitbucket introduced the possibility of parallelizing build steps 
(https://confluence.atlassian.com/bitbucket/configure-bitbucket-pipelines-yml-
792298910.html#Configurebitbucket-pipelines.yml-ci_parallel). Paralellizing our 
release and debug compiles and the style tests should reduce the total wall-clock 
time needed for the continuous integration tests.