RhinoFish - 3:28 pm on Apr 19, 2011 (gmt 0)

if you get too specific, your account will be filled with inactive and low volume long tail keywords, making the mountain of data to be analyzed geometrically larger.

you're example is short enough to allow dki in the title, but be careful with long tail terms whose length is over your insertion limits - ctr can suffer, and therefore other performance metrics.

semantically, the key differences can be managed with negatives, i'd suggest you experiment with Mod Broad w/ Negs to chase these somewhat long tail words, phrase seems like overkill here (widget example words make it difficult to know the specificity involved here). so i'd likely do:
+Cheap +Red +Widget with mined Negs
rather than a grunchload of phrase matches (that sill require mining Negs).

if you split these two things, keep in mind that the original is affected as well. pull things apart far enough (over segmentation) and you'll have insufficient data to make sound analytical decisions. granularity / discreteness is opposed by analytical and management resource limits. experience (aka data) will teach you the right balance for this account.