Context. Millisecond pulsars are very useful for determining the properties of neutron stars, for testing General Relativity and for detecting gravitational waves. However, the number of known millisecond pulsars is very small compared to that of ordinary pulsars. Aims. We conducted one blind and three targeted searches for millisecond and submillisecond pulsars at radio frequencies. Methods. The blind search was conducted within 3 • of the Galactic plane and at longitudes between 20 • and 110 • . It takes 22 073 pointings to cover this region, and 5487 different positions in the sky (i.e. 25% of the total) were actually observed. The first targeted search was aimed at Galactic globular clusters, the second one at 24 bright polarized and pointlike radiosources with steep spectra, and the third at 65 faint polarized and pointlike radiosources. The observations were conducted at the large radiotelescope of Nançay Observatory, at a frequency near 1400 MHz, the exact value depending on the backend. Two successive backends were used, first a VLBI S2 system, second a digital acquisition board and a PC with large storage capacity sampling the signal at 50 Mb/s on one bit, over a 24-MHz band and in one polarization. The bandwidth of acquisition of the second backend was later increased to 48 MHz and the sampling rate to 100 Mb/s. The survey used the three successive setups, with respective sensitivities of 3.5, 2.2, and 1.7 mJy. The targeted-search data were obtained with the third setup and reduced with a method based on the Hough transform, yielding a sensitivity of 0.9 mJy. The processing of the data was done in slightly differed time by soft-correlation in all cases. Results. No new short-period millisecond pulsars were discovered in the different searches. To better understand the null result of the blind survey, we estimate the probability of detecting one or more short-period pulsars among a given Galactic population of synthetic pulsars with our setup: 25% for the actual incomplete survey and 79% if we had completed the whole survey with a uniform nominal sensitivity of 1.7 mJy. The alternative of surveying a smaller, presumably more densely populated, region with a higher sensitivity would have a low return and would be impractical at a transit instrument. Conclusions. The null result of our blind survey is compatible with our present understanding of the Galactic population of millisecond pulsars. In particular, there does not seem to be a large population of fast-rotating millisecond pulsars.