As discovered in Issue #6841, connections to soft-upgraded databases are significantly slower than those to their hard-upgraded counterparts. To get an idea of just how severe the performance loss is in the former case, take a look at the results of the repetition tests run by myself (Xavion) and Rick Hillegas.
In the meantime, I noticed that there doesn't seem to be any mention in Derby's documentation about the loss of performance experienced when using soft-upgraded databases. The only thing anyone seems to care about is the loss of functionality that one can expect when using an older database.
In fact, the "Soft upgrade limitations" article doesn't even exist anymore in Derby releases above v10.9. Given how old (and therefore mature) Derby is, people like myself assume that its database format doesn't change much between versions. Due to this, it'd never occur to us that soft-upgrading could be so detrimental to performance.