I am using DESeq2 to find the DE genes in a data set with 21 replicates ( 21 + 21, paired-design). I get the DE genes but with very small log2 fold changes (less than +/- 0.5) , as you can see in the plot below:

> summary(resdds, 0.05)
out of 22025 with nonzero total read count
adjusted p-value < 0.05
LFC > 0 (up)     : 2479, 11%
LFC < 0 (down)   : 3127, 14%
outliers [1]     : 0, 0%
low counts [2]   : 854, 3.9%
(mean count < 4)

Unshrunken LFC:

I could see that the mean expression is good and the sample size is large. I would like to know if the results makes sense and are trustable.

P.S: The GO analysis looks good as well.

"I would like to know if the results makes sense and are trustable."

Yes they are. This is just a consequence of having true differences and sufficient read depth and sample size to detect them.

We discuss this in the section "Hypothesis tests with thresholds on effect size" of the DESeq2 paper.