1. technical problems in the assays making samples to load different (for some reason) in the different cells (i.e. in samples from cells treated with less or more TNF concentration) or
2. that indeed TNF alfa afects 18S expression (18S genes; I assume that indicates 18S expression).
I would vote for the first more than for the second option. Important data missing is time. To have putative differences in 18S expression you need enough time for changes of expression to occur.
There are a lot of papers out that dispute the value of 18S as a reference gene. It's important in any experiment to determine appropriate control for your cells and treatments. It is entirely possible that controls for one cell line need to be different with different treatment conditions since one treatment could be causing a change in energy storage resulting in GAPDH being inappropriate while another alters cell morphology making actin inappropriate. 18S could be inappropriate if there is a change in proliferation rate or total protein synthesis. Also some experiments (PCR with control and GOI in a single well) may fail with 18S because it tends to be orders of magnitude greater in abundance than the target gene; as such, 18S has different kinetics than the other gene and may exhaust reagents before the other gene is amplified.