TTest significance in replicate trials

I'm having difficulty interpreting the results of two consecutive
trials that were analyzed with a simple T-test assuming equal
variances. Two samples with N=8 in each sample were used, and the mean
of set A was found to be different from the mean of set B with a
p-value of 0.016. Due to some problems with the way the experiment was
conducted the first time, we decided to re-run the test with a new set
of samples. Again, the ttest rejected the null hypothesis with p-value
of 0.011. The problem is, this time the values of the means switched
in rank - that is, in the first trial A > B; whereas in the second
trial, B > A. The reasons for repeating the trial should not have
affected the results in this way (the analyst forgot to record some
bureaucratically imporant experiment details). Can anyone offer
guidance in explaining how the results indicate the samples are truly
different? Shouldn't the direction of the mean difference be
consistent?

.