I had doubts, but not a second device to do some tests, until my father bought himself an Oregon Scientific BAR206 radio-controlled clock and weather station, like mine. He noticed a difference of +0.5°C in the measurement of indoor temperature between BAR206 (production code 11A13 ?) and his old station.
So we gathered all his external sensors in one place and waited for readings to settle: BAR206 internal sensor was 0.5 °C higher. One value against three values.
I repeated the test at home with my BAR206 (production code 05A11 ?) and I could confirm the extra half-degree.
I think the reading error is due either to a firmware bug or the tight case that doesn't let air to flow inside.
I will now remove the back cover and see what happens. Check back for results.