This is not my understanding. It is completely dependent on the
threshold used, which can be arbitrary. See Supplemental figure 7, page
16 of this document:
http://www.jneurosci.org/cgi/data/30/6/2268/DC1/1
This was the motivation for Smith & Nichols' TFCE:
http://www.ncbi.nlm.nih.gov/pubmed/1
It should be the case that the clusters in the 2.3 should be clusters in
the 1.3, but there p-values will be different. It is possible that a
cluster that is sig in the 2.3 is not sig in the 1.3, particularly small
clusters like the two you have shown. Small clusters may be very likely
at 1.3 u
