The increasing use of bibliometric indicators in science policy calls for a reassessment of their robustness and limits. The
perimeter of journal inclusion within ISI databases will determine variations in the classic bibliometric indicators used
for international comparison, such as world shares of publications or relative impacts. We show in this article that when
this perimeter is adjusted using a natural criterion for inclusion of journals, the journal impact, the variation of the most
common country indicators (publication and citation shares; relative impacts) with the perimeter chosen depends on two phenomena.
The first one is a bibliometric regularity rooted in the main features of competition in the open space of science, that can
be modeled by bibliometric laws, the parameters of which are “coverage-independent” indicators. But this regularity is obscured
for many countries by a second phenomenon, the presence of a sub-population of journals that does not reflect the same international
openness, the nationally-oriented journals. As a result indicators based on standard SCI or SCISearch perimeters are jeopardized
to a certain extent by this sub-population which creates large irregularities. These irregularities often lead to an over-estimation
of share and an under-estimation of the impact, for countries with national editorial tradition, while the impact of a few
mainstream countries arguably benefits from the presence of this sub-population.