The biggest difference between the 2 queries is the second one is really a JOIN, not a Left Outer Join. By adding the second WHERE condition on a property in pcu you are making this a JOIN and allowing the Optimizer to start in either table. I created the tables on my system and here is what I am seeing
The first query starts in ps using the item index and then joins to the pcu table using a temp file built from the master map (same as the plan you show)
The second query is starting with the pcu table and the doing a join to the ps table using a temp file build from the Item index and the master map.
Why don't we use the Item index in pcu? I can't tell you exactly why, but I know the 2 biggest factors we are looking at are the size of the maps on disk and the Selectivity of the properties involved. In both cases, we are going to need to read the master map so the question is this: Will reading in the Index map reduce the number of blocks we need to read of the master map to make it worth the effort? Without proper values generated by TuneTable, or supplied by you, the optimizer is left to guess at what these values are. In the first case, it has decided that reading the index will not reduce the number of block reads of the master map so it is not using it. This generally happens when we think the Master Map is very small.
To take this any further I would need you to run TuneTable on both of these tables and see if the poor plan is still used. If it is still used please provide the storage info for both classes so I can investigate this.