Cuffdiff Returns Small P-Value for Conditions with Nearly Identical FPKM Values

Question

[bdiciac2]

I was using the tuxedo package to analyze RNAseq data. Tophat to map reads, cufflinks to estimate abundances, cuffmerge to merge files, and cuffdiff to determine DEGs. I have several questions on the cuffdiff output, in which I used version 2.2.1

#gene.exp output
gene                sample_1    sample_2    status  value_1 value_2  log2.fold_change.  test_stat   p_value q_value
Osm6                Kc167_NT60  Kc167_XR60  OK      11.0958 10.967   -0.016852         -0.00819555  0.0163  0.91993
CG4297              Kc167_NT60  Kc167_XR60  OK      3.28251 13.4614   2.03596           0.197429    0.0367  0.91993
Mst57Db             Kc167_NT60  Kc167_XR60  OK      0       6.35973   Inf               NA          0.0084  0.91993
snoRNA:CG16892-a    Kc167_NT60  Kc167_XR60  OK      0       18.2782   Inf               NA          0.02785 0.91993

#Replicate FPKM Values for gene Osm6 (from genes.read_group_tracking)
gene    condition   replicate   raw_frags   internal_scaled_frags   external_scaled_frags   FPKM    effective_length    status
Osm6    Kc167_NT60  0           296         289.701                 289.701                 8.75187      -              OK
Osm6    Kc167_NT60  1           529         467.688                 467.688                 13.4397      -              OK
Osm6    Kc167_XR60  0           268         285.888                 285.888                 8.43789      -              OK
Osm6    Kc167_XR60  1           441         471.429                 471.429                  13.496      -              OK

1. Why does gene Osm6 give a p-value of 0.02 even though the magnitude of the test statistic is very small and the FPKM values between treatments are almost identical (11.10 vs 10.97)? I would have expected a much larger p-value.
2. How is a p_value calculated for genes with "NA" test statistics?

Thank you