Why is in Qt/Opengl glMapBuffer slower than glBufferData?

PowerNow

I tried on Android for a huge terrain model (around 10Mill vertices/indices, 250MB)

m_vertexBuf.allocate(m_pVertices->size() * sizeof(QVector4D));
void* buffer_data = m_vertexBuf.map(QOpenGLBuffer::WriteOnly);
memcpy(buffer_data, m_pVertices->constData(), m_pVertices->size() * sizeof(QVector4D));
m_vertexBuf.unmap();
m_vertexBuf.release();

instead of

m_vertexBuf.allocate(m_pVertices->constData(), m_pVertices->size() * sizeof(QVector4D));

but it's 0.4sec slower(!) during loading.