Currently jumbo chunks are ignored from the balancing equation altogether.

This means that if one shard has a large number of jumbo chunks it will achieve equilibrium with the other shards on the number of non-jumbo chunks and have the additional jumbo chunks as overhead.

Eg.
Collection has 334 Chunks of which 70 are jumbo.
Shard1(primary shard) has 158 chunks (88 + 70)
Shard2 has 88 chunks
Shard3 has 88 chunks

The balancer sees all three shards as having 88 chunks as the 70 jumbo chunks are ignored from the count. This leaves the cluster heavily unbalanced.