A second mention of CLEAN stripes. (Cornwell 1983) suggested that this instability is suppressed by the addition of a delta function to the dirty beam. For obvious reasons, this is popularly termed the `Prussian helmet' CLEAN. The spike is usually set at about 0.3 (see EXPLAIN APCLN). Together with this, one usually makes the loop gain smaller, to perhaps 0.05 or even 0.01. However, recent wisdom suggests that the Prussian helmet is probably a red herring. I suggest that rather than mixing with teutonic hats, you use SDCLN, which takes a more reasonable approach to CLEAN stripes.
The EXPLAIN files for APCLN, SDCLN, and MX contain lots of information about CLEAN, should you like to read some more.