When running any workflow with Enable Cache Features turned on, even though the workflow may not use any spatial transformers, if the reader detects a set of coordinates, FME automatically builds spatial indexes for each transformer the data passes through. This slows down the workflow considerably and can cause it to crash if large volumes of data are being processed.
One solution (as suggested by 1Spatial) is to manually check the readers to ensure that any field that may been detected as a coordinate is changed to a different type, e.g. varchar, before running the workflow. This works but is labourious.
Alternatively, and my suggestion, wouldn’t it be simpler for FME to add a switch to allow spatial indexing to be turned off when not required?