Reading ArcASCII / ArcGRID doesn't handle NODATA value correctly - or I don't understand the concept

Since I switched Geopublisher and AtlasStyler from GT 2.3 to GT 2.6 a long time ago, we always had problems with NODATA and transparency in ArcASCII files. Our quick solution at that time was to continue to use the old ArcGridRaster class which worked fine to create GridCoverages. My plan for GP 1.6 is to used only readers, so I looked into the problem again.

I wrote two tests that load an ArcASCII dataset (SWAN_NURC_LigurianSeaL07_HSIGN.asc) and look at it's NODATA value. My understanding is, that when the "coverage/raster/image driver" is aware of a NODATA value being defined, that that NODATA value should not appear in the coverage. The tests show that the NODATA value appears in the coverage and in the image just like a normal value.

In my understanding, NODATA values are workarounds for data-formats that do not support null values. Once they are identified, they should be replaced with null or NaN so that any statistics or styling done on the data ignores them.

Please, can someone look at these two tests and tell me whether I found two bugs or whether I just do not understanding the concept of NODATA values correctly. Thanks.